Purpose: Develop custom phantoms with variable, controllable, and repeatable textures for the assessment of high-resolution CT.
Methods: Procedural routines for texture generation were developed based on constrained sphere packing within specified volumes. Spheres define voids within 3D-printed phantoms and control of sphere sizes and degree of intersection permitted a range of possible textures. Repeatability in phantom production was investigated by printing ensembles phantoms of the same design. Phantoms were scanned with high-resolution cone-beam CT and registered, permitting computation of standard deviation volumes to quantify variability. Various printing conditions were explored including hole sizes (2.0, 1.0, and 0.5 mm) and different Fused Deposition Modeling (FDM) print nozzles (0.4 mm and 0.2 mm). Texture parameters were adjusted to mimic human tissue textures, and variable contrast was achieved via immersion in a potassium phosphate solution. Radiomics measures (trabecular thickness, spacing, and bone volume fraction) of six bone-mimicking samples were computed and compared to the digital design.
Results: Smaller hole sizes were not resolved in 3D prints but still resulted in high-resolution textures. The textures further depended on FDM nozzle size. Higher variability was found for phantoms produced with the smaller 0.2 mm nozzle. With appropriate hole size and overlap, anthropomorphic textures mimicking trabecular bone were generated. For radiomics metrics, the trabecular thickness was systematically overestimated by 26%, spacing was underestimated by 10%, and bone volume was within 6%. While these measures were biased, standard deviations of these values were low, indicating high repeatability. Texture contrast was successfully controlled via immersion in a potassium phosphate solution.
Conclusion: The 3D printed textured phantoms offer a highly repeatable method to probe the ability of high-resolution CT to reproduce textures in reconstructed images. With increasing focus on task-based image quality and radiomics, such custom phantoms have the potential to play an increasing role in imaging performance assessments.
Funding Support, Disclosures, and Conflict of Interest: This work was supported, in part, by NIH grants R01EB025829 and R21CA219608.